Forest monitoring

Global Forest Change gives every 30-metre pixel on Earth a tree cover in 2000 and a year of loss since, and it has made forest statistics look easy. In a 3,000 km² window of Rondônia the headline is stark: 44% of the forest standing in 2000 was gone by 2024. Every word of that sentence is a decision — forest means canopy above 30%, the pixels are 0.00025° and shrink with latitude, loss means any stand-replacing event, and the figure belongs to one release of the data.

Change any of them and the number moves. A 10% threshold gives 41%, a 75% threshold 49%. The common "0.09 ha per pixel" overstates area by 19% at 10°S and 145% at 61°N. Protected land a kilometre inside the boundary lost 7.7% of its forest against 58% a kilometre outside, but comparing the two sides still mixes protection with access, and the boundary band made the estimate worse rather than better. A Sentinel-2 detector disagreed with the annual loss map almost entirely, because the two measured different intervals.

The concepts below settle what the data measures, how forest is defined, how rates, fragmentation and avoided loss are computed; the how-to guides work through real Hansen tiles, OSM protected areas and roads, H3 hotspots, land cover products, canopy height and Sentinel-2; and the fixes are the five ways forest numbers go wrong while looking authoritative.

Learning paths run through this topic too — see the end of this page.